Math-Mind-AI

• 

• 

  Backend

• 

  Solution

Agentic-RAG architecture system that replicates a mathematical professor. When presented with a mathematical question, the system should understand and generate a step-by-step solution, simplifying it for the student. The response must first check the existing knowledge base for relevant information. If the solution is not found in the knowledge base, the system should perform a web search and then generate the step-by-step solution. The Math-agent must have a solid routing pipeline: AI Gateway mainly guardrails Integrate the routing agent with AI gateway by adding the Input and Output guardrails. Do proper research on how the guardrails is incorporated. The app should be focused on only delivering education based content mainly: Mathematics.

Knowledge Base Creation To build an Agentic RAG, the system must adapt to a knowledge base to retrieve relevant questions effectively. You can choose any dataset or math module as the knowledge base and store it in a VectorDB. If the user’s question exists in the knowledge base, the system should retrieve it and generate a step-by-step solution to provide simplified results.

Web Search or Using MCP If the user’s question is not found in the knowledge base, the system should perform a web search to fetch the result and generate a response. However, if the question is not available online, the system must ensure it does not provide incorrect results. Furthermore, for this step, you need to have a pipeline for Web Search extraction. Usage of Model Context Protocol (MCP) is MUST.

Human-in-the loop mechanism Incorporate an evaluation or feedback agent layer to enhance the agent's performance through self-learning capabilities. Note: You need to take human feedback and refine the response accordingly. The final response is subjective and requires a Human-in-the-Loop mechanism for feedback and validation. Bonus: If you use the DSPy library for this.

Output:- Input & Output guardrails used for privacy. What approach did you take and why? Knowledge Base - Dataset used and details on it: including 2-3 questions to try the system. Web Search capabilities or MCP Setup: - include 2-3 questions that are not from the knowledge base. Need to include the strategy taken for Web extraction or MCP Setup. A detailed report on the Human-in-a-loop routing for the Agentic workflow taken to build the Math Agent.

Built With

• agent
• ai
• autogen
• dspy
• fastapi
• fine-tuning
• flask
• gemini
• groq
• langchain
• mcp
• python
• qdrant
• rag
• react
• typescript